Paukkunen et al. BioMedical Engineering OnLine (2015) 14:16 DOI 10.1186/s12938-015-0013-9

RESEARCH Open Access

Unified frame of reference improves inter-subjectvariability of seismocardiogramsMikko Paukkunen1*, Petteri Parkkila1, Raimo Kettunen2 and Raimo Sepponen1,3

* Correspondence:mikko.paukkunen@aalto.fi1Department of ElectricalEngineering and Automation, AaltoUniversity, P.O. BOX: FI-13340, 00076Helsinki, FinlandFull list of author information isavailable at the end of the article

Abstract

Background: Seismocardiography is the noninvasive measurement of cardiacvibrations transmitted to the chest wall by the heart during its movement. Whilemost applications for seismocardiography are based on unidirectional accelerationmeasurement, several studies have highlighted the importance of three-dimensionalmeasurements in cardiac vibration studies. One of the main challenges in usingthree-dimensional measurements in seismocardiography is the significant inter-subjectvariability of waveforms. This study investigates the feasibility of using a unified frameof reference to improve the inter-subject variability of seismocardiographic waveforms.

Methods: Three-dimensional seismocardiography signals were acquired from tenhealthy subjects to test the feasibility of the present method for improving inter-subjectvariability of three-dimensional seismocardiograms. The first frame of referencecandidate was the orientation of the line connecting the points representing mitralvalve closure and aortic valve opening in seismocardiograms. The second candidatewas the orientation of the line connecting the two most distant points in the threedimensional seismocardiogram. The unification of the frame of reference wasperformed by rotating each subject’s three-dimensional seismocardiograms so that thelines connecting the desired features were parallel between subjects.

Results: The morphology of the three-dimensional seismocardiograms varied stronglyfrom subject to subject. Fixing the frame of reference to the line connecting the MCand AO peaks enhanced the correlation between the subjects in the y axis from0.42 ± 0.30 to 0.83 ± 0.14. The mean correlation calculated from all axes increased from0.56 ± 0.26 to 0.71 ± 0.24 using the line connecting the mitral valve closure and aorticvalve opening as the frame of reference. When the line connecting the two mostdistant points was used as a frame of reference, the correlation improved to 0.60 ± 0.22.

Conclusions: The results indicate that using a unified frame of reference is a promisingmethod for improving the inter-subject variability of three-dimensional seismocardiograms.Also, it is observed that three-dimensional seismocardiograms seem to have latentinter-subject similarities, which are feasible to be revealed. Because the projections ofthe cardiac vibrations on the measurement axes differ significantly, it seems obligatoryto use three-dimensional measurements when seismocardiogram analysis is based onwaveform morphology.

Keywords: Seismocardiography, Ballistocardiography, Aaccelerometer, Three-dimensional,Inter-subject variability, Frame of reference

© 2015 Paukkunen et al.; licensee BioMed Central. This is an Open Access article distributed under the terms of the CreativeCommons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, andreproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedicationwaiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwisestated.

Paukkunen et al. BioMedical Engineering OnLine (2015) 14:16 Page 2 of 9

BackgroundSeismocardiography (SCG) is the noninvasive measurement of cardiac vibrations trans-

mitted to the chest wall by the heart during its movement [1]. The emergence of SCG

measurement can be traced back to 19th century when Gordon reported observing a

heartbeat while standing on a scale [2]. Gordon’s observation led off the development

of several measurement methods for cardiac induced vibrations such as ballistocardio-

graphy (BCG). BCG is a method in which the reaction forces acting on the whole hu-

man body are measured. Thus, SCG can be viewed as a derivative of BCG. The

emergence of SCG as a separate measurement technique dates back to 1960s when the

pioneering works of Baevsky and Bozhenko [3,4] were published. The advances in elec-

tronic accelerometer technology have resulted in multiple applications for SCG, such

as ischemic heart disease detection, gating of cardiac imaging and therapy, as well as

smartphone-based heart activity monitoring [5-9].

Most applications for SCG are based on unidirectional acceleration measurement,

usually done in the back-to-front direction. However, several studies have highlighted

the importance of three-dimensional (3-D) measurements in cardiac vibration studies

[10-13]. For example, Migeotte et al. concluded that the vibration waveforms observed

on the foot-to-head axis are poorly correlated to the magnitude of the maximum sys-

tolic force vector computed using all three axes of acceleration [10]. This finding indi-

cates that the vibrations of the heart might be projected disproportionately between

the three orthogonal axes of measurement and may thus lead to misinterpretation of

the waveforms and the underlying physiological phenomena. Another fairly recent

paper reported a method that was based on 3-D SCG vector trajectories which were

used for computing the IJ-waveform amplitudes [14]. The IJ-waveforms were proposed

to be related to cardiac output. The report found that that since the 3-D approach

takes into account the uneven distribution of vibrations between the measured axes,

the achieved results might be independent from slight misalignments of the SCG sen-

sor. This indicates that if slight misorientations of the SCG sensor are accountable, the

compensation of inter-subject variation due to physiological differences should also be

feasible.

Inter-subject variability of seismocardiograms, which hinders the deployment of

SCG in to clinical practice, is a widely observed phenomenon in SCG studies [15,16].

We hypothesize that a unified frame of reference would facilitate improvement of

inter-subject variability of seismocardiograms. The hypothesis relies on two assump-

tions. The first assumption is that asymptomatic hearts function similarly in general.

Thus, only modest inter-subject variability should be seen when measuring SCG from

healthy subjects. However, significant inter-subject differences in seismocardiograms

are continuously reported even in healthy subjects. This indicates that SCG measure-

ments are prone to variability due to other factors than differences in cardiac function.

These other factors could be anatomical differences such as the orientation of the

heart and the aortic arc or differences in the mechanical coupling of the myocardial

vibrations to the sternum. The second assumption is that 3-D acceleration measure-

ment captures the vibrations of the chest sternum in every direction. Thus, even if the

vibrations of the heart are projected differently between subjects, the use of a unified

frame of reference might straighten the vector trajectories and reveal latent similarity

in the seismocardiograms.

Paukkunen et al. BioMedical Engineering OnLine (2015) 14:16 Page 3 of 9

This study investigates the feasibility of using a unified frame of reference to improve

the inter-subject variability of SCG waveforms. To test the feasibility of the proposed

method, a data set from ten subjects is acquired and pre-processed.

MethodsEquipment

The 3-D SCG signals were measured with three orthogonally mounted accelerometers

(SCA610-C21H1A, Murata Electronics, Finland). The accelerometers and associated elec-

tronics are both described in detail in the authors’ previous work [17]. For this study, we

mounted the individual accelerometers to a smaller package than before using eyesight to

orient the sensors. Due to the mounting process, the orthogonality of the acceleration

sensors could not be guaranteed. The largest orientation error between the accelerometers

was measured between the xy plane and the z axis (5 degrees). Thus, in the worst case,

any single axis of acceleration measurement will include an error of 9% compared to

ideally orthogonal accelerometers. The acceleration sensors had a sensitivity of 2 V/g, true

DC response and rated output noise of approximately 60 μgrms in the frequency band of 1

to 50 Hz. Prior to use, the accelerometers were tested at VTI’s (now Murata Electronics)

laboratories where they were shown to have their -3 dB points at 47 Hz. The acceleration

signals were anti-alias filtered with 8th-order Bessel lowpass filters with a rated cut-off fre-

quency of 100 Hz and attenuation of at least 96 dB at 800 Hz. The filters are coupled in

the Sallen-Key topology using quad operational amplifiers (AD8630, Analog Devices,

USA). For this particular study, the gain stage of the acceleration measurement in [17] is

bypassed, resulting in a DC response and a gain of 0 dB for the acceleration measurement.

Three individual accelerometers were used because, to the best of the authors’ knowledge,

no integrated options with similar characteristics (high sensitivity, low noise, DC re-

sponse) were available during the time of implementation. In terms of accuracy, there is

no inherent benefit in using individual accelerometers compared to integrated 3-D accel-

erometers. In particular, recent research has applied integrated 3-D accelerometers with

similar noise and sensitivity performance to SCG but with no DC response [18]. A DC re-

sponse would be important in other SCG applications where the instantaneous inclination

of the sternum was of interest.

The electrocardiography (ECG) lead II was measured using a commercial wireless

dual-lead ECG system (BC-ECG2, BIOPAC Systems Inc, US). Respiratory efforts were

detect using a commercial wireless respiratory effort detection system (BN-RESP-

XDCR, BIOPAC Systems Inc, US). All data were captured using the MP150 Data

Acquisition System (BIOPAC Systems Inc, US). The acceleration signals were coupled

to the MP150 through a Universal Interface Module (UIM100C, BIOPAC Systems Inc,

US). Preprocessing of the signals was done in the Acknowledgment environment

(BIOPAC Systems Inc, US). All post-processing of the data was done in the Matlab

environment (MATLAB R2013b, Mathworks, US).

In vivo measurements

Human subject protocol

The present study was completely conducted in the premises of Aalto University,

Espoo, Finland. Ten male volunteers with an average age of 33 years (standard

Paukkunen et al. BioMedical Engineering OnLine (2015) 14:16 Page 4 of 9

deviation (SD) 8.5 years), an average weight of 79.5 kg (SD 14.9 kg) and an average

height of 175.2 cm (SD 7.1 cm) were measured as test subjects. During the measure-

ments, the subjects were at resting supine position. To minimize breathing-induced

variability, the subjects were instructed to hold their breath as long as comfortable.

The accelerometer package was positioned on the lower part of the sternum about

one centimeter above the xiphoid process using double-sided adhesive tape. The accel-

erometer package was mounted so that one accelerometer measured the acceleration in

the back-to-front (z axis), one in the right-to-left (x axis), and one in the foot-to-head

direction (y axis) (see Figure 1).

The contents of the test and the course of its events were explained individually to

the subjects before the study. In addition, a written consent was received. The study

did not contain any such intervention in the physical integrity of the test subjects, or

any other features that would require an ethical review as considered by the National

Advisory Board on Research Ethics in Finland.

Selecting the frame of reference

The rationale behind using a unified frame of reference lies in the assumption that

most hearts function similarly in general. Thus, the inter-subject differences might be

mainly due to anatomical differences and differences in the mechanical coupling of the

heart’s vibration to the sternum. Some methods that might be useful in managing the

variability of seismocardiograms have been presented in the literature. The approaches

include the division of heartbeats in to inspiration and expiration heartbeats [19] and

normalization by resampling each heartbeat to be of same length [13], for example.

Castiglioni et al., for example, proposed that the magnitude of the IJ-waveform com-

puted from 3-D seismocardiograms is independent of slight orientation differences of

Figure 1 Measurement axes. The x axis is the left-to-right direction, the y axis the foot-to-head directionand the z axis the back-to-front direction.

Paukkunen et al. BioMedical Engineering OnLine (2015) 14:16 Page 5 of 9

the SCG sensor [14]. However, the use of orientation information, which is readily ob-

tainable from 3-D SCG signal, is rarely reported.

The authors hypothesize that if some orientation-related representative feature could

be extracted from every seismocardiogram, this feature could be used to normalize the

orientation of seismocardiograms so that each seismocardiogram could be interpreted

in the same frame of reference. Thus, this approach would facilitate reducing the effect

of mechanical coupling and anatomic differences in the seismocardiograms. Two candi-

dates were selected to serve as frames of reference. The first candidate was the orienta-

tion of the line connecting the points that represent mitral valve closure (MC) and

aortic valve opening (AO) (see Figure 2). During this time interval, the ventricles con-

tract with no change in volume (i.e. isovolumic contraction) and no blood flows to the

aorta. Thus, only respiration efforts and the movement of the heart affect the SCG

waveforms. As the baseline shifts in SCG signal due to respiratory efforts (i.e. the in-

clination of the sternum changing) can be effectively eliminated with high-pass filtering,

the orientation of the filtered SCG signal during isovolumic contraction can be as-

sumed to depict the orientation of the heart’s mechanical axis. Recent research has sug-

gested that breathing manifests also in SCG morphology [20]. In this study, we attempt

to minimize these changes by restricting the analysis to epochs of breath holding. The

second candidate was the orientation of the line connecting the two most distant points

in the 3-D seismocardiogram (see Figure 2). This line was assumed to represent the

direction in which the heart could produce maximal systolic force.

Signal processing

Pre-processing

All SCG and ECG signals were digitally filtered in BIOPAC with a band-pass filter

using a 0.5 Hz lower and a 40 Hz upper cut-off frequency. Based on the respiratory

Figure 2 Frames of reference. The red asterisk marks the MC point and the green asterisk the AO point.The green and red circles are the two most distant points in the seismocardiogram. The red dashed lineconnects the MC and AO points, and the green dashed line connects the two most distant points in theseismocardiogram. The units are in milli-gs (1 mg = 9,81 mm/s2).

Paukkunen et al. BioMedical Engineering OnLine (2015) 14:16 Page 6 of 9

signal, the data where the breath hold takes place was manually selected for each sub-

ject. R peaks were detected using a simplified Pan-Tompkins algorithm [21] by first

finding all cardiac cycles with an arbitrary length and then seeking the maximum values

inside each cycle. Extraction of SCG cycles was then performed by segmenting the sig-

nals to 700-ms long windows starting 100 ms before and ending 600 ms after each

ECG R peak. We found that the timings of the relevant SCG features in relation to the

R peaks were not associated with variable heart rate, and thus normalization would

have drastically affected the averaging process. Therefore, in this study, no heart rate

normalization was performed. After segmentation, ensemble averaging was done by

using the largest number of heartbeats that was available from all subjects (11 beats).

These heartbeats were selected by minimizing the mean standard deviation of the

resulting ensemble averages. MC and AO peaks were manually detected from averaged

z axis SCG signals using the annotation scheme proposed by Crow et al. as a guideline

[22]. The MC point was identified as the positive peak on the z-axis seismocardiogram

following the onset of the QRS complex, whereas the AO point is the positive peak

after the MC point.

Computing the frame of reference

A 3-D seismocardiogram is defined as a vector function of time with magnitude offfiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiX2 þ Y 2 þ Z2

pwhere each of the SCG axes represents a Cartesian coordinate in 3-D

space. Ensemble averaged 3-D seismocardiograms for each subject were rotated sepa-

rately based on the two frames of reference, which were the line connecting the most

distant points (Frame 1) and the line connecting the MC and AO peaks (Frame 2), as

described previously. To unify the orientation of the original 3-D seismocardiograms,

they were rotated so that either frame 1 or frame 2 was parallel with the z axis

(Figure 3). Rotation was performed in a plane with a normal vector described by the

cross product of the frame of reference and the z axis.

Evaluation of the inter-subject variability of seismocardiograms

Squared Pearson correlation coefficients (R2) as well as root-mean-square error (RMSE)

and normalised root-mean-square error (NRMSE) in the MC-AO interval for each

SCG axis were calculated in order to quantify the inter-subject variability in the 3-D

seismocardiograms both before and after the rotations. For each SCG axis, this pro-

cedure resulted inX10 − 1

1

n ¼ 45 different correlation coefficients, RMSE, and NRMSE

values between the ten subjects.

Results and discussionThe effect of rotations on the inter-subject correlation, RMSE, and NRMSE is shown in

Table 1. The morphology of the 3-D SCG waveforms varied strongly from subject to

subject, as evidenced by the relatively weak inter-subject correlation (mean correlation

of 0.39 on the x axis, 0.42 on the y axis, and 0.86 on the z axis) of the unprocessed 3-D

seismocardiograms. Fixing the frame of reference to the line connecting the MC and

AO peaks (i.e., frame 2) considerably enhanced the correlation between the subjects in

the y axis (mean correlation of 0.52 and 0.83 using frame 1 and frame 2, respectively).

Figure 3 Rotation of seismocardiograms. The leftmost column shows the original unrotated signal of asubject, the plots in the middle show the original signal rotated so that the line connecting the two mostdistant points is parallel to the z axis (frame 1), and the rightmost column shows the original signal rotatedso that the line connecting the MC and AO points is parallel to the z axis (frame 2). The units are in mg.

Paukkunen et al. BioMedical Engineering OnLine (2015) 14:16 Page 7 of 9

Also, the NRMSE decreased from 0.43 to 0.29 on using frame 2, suggesting the rotation

process decreased the inter-subject variability. However, the impact of using a unified

frame of reference was not as prominent in the other axes. The x axis correlations were

only slightly improved and RMSE slightly decreased using either frame of reference.

The proposed method did not significantly affect the standard SCG z-axis features.

This is quantified by the high correlation between the original and rotated z axis signals

(subject-wise mean of 0.99 for frame 1 and 0.95 for frame 2) as well as the consistency

of the original and rotated z axis waveform RMSEs. Also, it can be visually confirmed

from Figure 4 that the z-axis waveforms remain similar despite the rotation process.

Using the frame 2, the mean correlation across all axes was improved from 0.56 to

0.71. Also, 44 out of 45 y axis correlation coefficients were improved using frame 2.

Despite the seemingly low amount of increased coefficients in X and Z axes, well over

half of the coefficients in separate axes did improve when both frames of reference

were considered (31, 44 and 27 in the x, y, and z axis, respectively).

Although the y axis correlations were improved remarkably when a unified frame of

reference was used, the x axis correlations remained weak. This suggests that the x axis

does not contain significant physiological information. As the accelerometers used in

this study were very sensitive (2 V/g), it seems that insufficient coupling of cardiac and

blood flow-induced vibrations to the sternum might partly cause the weak inter-subject

correlation. Migeotte et al. also observed that most of the time the cardiac function is

Table 1 Effect of rotation

RMSE [mg] Normalized RMSE Correlation

Axis x y z x y z x y z

Unrotated 4.91±1.76 3.63±1.59 7.26±3.40 0.43±0.12 0.43±0.17 0.13±0.05 0.39±0.30 0.42±0.31 0.86±0.13

Frame 1 3.90±1.56 3.62±1.81 7.47±3.51 0.55±0.19 0.45±0.16 0.13±0.05 0.44±0.33 0.52±0.31 0.85±0.13

Frame 2 4.40±1.68 4.60±2.49 7.27±3.63 0.47±0.21 0.29±0.13 0.14±0.05 0.43±0.31 0.83±0.14 0.85±0.14

Figure 4 Effect of rotation on individual seismocardiogram trajectories. The individual seismocardiogramtrajectories without rotation (left), with rotation using frame 1 (center), and with rotation using frame 2 (right)are shown. The gray plots are the mean seismocardiograms of individual subjects. The blue lines are the meanof the means of seismocardiograms of individual subjects.

Paukkunen et al. BioMedical Engineering OnLine (2015) 14:16 Page 8 of 9

projected almost entirely on the yz-plane [13]. To enhance 3-D SCG, the characteristics

of the x axis should be further studied.

While the present method was feasible to remarkably improve the y axis correlations,

the z axis correlations were virtually unaffected. The possibility of masking or modifica-

tion of individual biological issues cannot, however, be excluded and it is yet unclear

whether these biological issues are important. The proposed method does not wipe out

information but only displays it differently. The mounting of the accelerometers might

have affected the results. As the orientation error found in the accelerometers was rela-

tively small, the effect of this error might be negligible. However, the degree of the

effect must be quantified in future research. Also, we suggest that in future work

the accelerometers should be mounted using a repeatable process that will guarantee

the orthogonality of the accelerometers.

The z axis SCG has been proposed as feasible to be used in cardiac time interval

measurements [23]. Given the high z axis correlations, it seems that the capability of

the z axis to be used in measuring cardiac time intervals was not affected when a uni-

fied frame of reference was used. This observation further asserts the feasibility of using

a unified frame of reference to improve 3-D SCG measurements.

ConclusionsThe results achieved in this study indicate that the use of a unified frame of reference is a

promising method for improving the inter-subject variability of 3-D seismocardiograms.

A larger sample size is needed to reassert the validity of this method. More importantly,

these results show that 3-D seismocardiograms have latent inter-subject similarities,

which are feasible to be revealed. Because the projections of the cardiac vibrations on the

measurement axes differ greatly, it seems obligatory to use 3-D SCG measurements when

analysis is based on SCG waveform morphology rather than time intervals.

Competing interestsThe authors declare that they have no competing interests.

Authors’ contributionsMP designed the study, drafted the manuscript, and participated in data analysis. PP carried out the data analysis andparticipated in drafting the manuscript. RK and RS participated in designing the study and critically revised themanuscript. All authors read and approved the final manuscript.

Paukkunen et al. BioMedical Engineering OnLine (2015) 14:16 Page 9 of 9

AcknowledgementsMikko Paukkunen is funded by The Finnish Cultural Foundation and The Finnish Society of Electronics Engineers.

Author details1Department of Electrical Engineering and Automation, Aalto University, P.O. BOX: FI-13340, 00076 Helsinki, Finland.2School of Medicine, University of Eastern Finland, Kuopio, Finland. 3Health Factory, Aalto University,P.O. BOX: FI-13340, 00076 Helsinki, Finland.

Received: 8 August 2014 Accepted: 10 February 2015

References

1. Salerno D, Zanetti J. Seismocardiography: a new technique for recording cardiac vibrations. Concept, method, and

initial observations. J Cardiovasc Technol. 1990;9((2):111–8.2. Gordon J. Certain molar movements of the human body produced by the circulation of the blood. J Anat Physiol.

1877;11(Pt 3):533–6.3. Bozhenko BS. Seismocardiography–a new method in the study of functional conditions of the heart. Ter Arkh.

1961;33:55–64.4. Baevsky R, Egorov A, Kazarian L. Metodika seismokardiografii. Kardiologia. 1964;18:87–9.5. Dinh A. Heart activity monitoring on Smartphone. Int Proc Chem Biol Environ Eng. 2011;11(9):45–9.6. Korzeniowska-Kubacka I, Bilińska M, Piotrowicz R. Usefulness of seismocardiography for the diagnosis of ischemia

in patients with coronary artery disease. Ann Noninvasive Electrocardiol. 2005;10(3):281–7.7. Wick C, Su J, McClellan J, Brand O, Bhatti P, Buice A, et al. A system for seismocardiography-based identification of

quiescent heart phases: implications for cardiac imaging. IEEE Trans Inf Technol Biomed. 2012;16(5):869–77.8. Tavakolian K, Khosrow-Khavar F, Kajbafzadeh B, Marzencki M, Blaber A, Kaminska B. Precordial acceleration signals

improve the performance of diastolic timed vibrations. Med Eng Phys. 2013;35(8):1133–40.9. Becker M, Roehl A, Siekmann U, Koch A, de la Fuente M, Roissant R, et al: Simplified detection of myocardial

ischemia by seismocardiography. Herz 2013, [Epub ahead of print]:1–7.10. Migeotte P, De Ridder S, Tank J, Nathalie Pattyn, I I Funtova, Roman M Baevsky, et al.: Three dimensional

ballisto-and seismo-cardiography: HIJ wave amplitudes are poorly correlated to maximal systolic force vector.In Proceedings of the Engineering in Medicine and Biology Society (EMBC), 2012 Annual International Conferenceof the IEEE: 28 August - 1 September 2012; San Diego, California, USA. 2012.

11. McKay WPS, Gregson PH, McKay BWS, Militzer J. Sternal acceleration ballistocardiography and arterial pressurewave analysis to determine stroke volume. Clin Invest Med. 1999;22(1):4–14.

12. De Ridder S, Migeotte P F, Neyt X, Pattyn N and Prisk G: Three-dimensional ballistocardiography in microgravity:A review of past research. In Proceedings of the Engineering in Medicine and Biology Society,EMBC, 2011 AnnualInternational Conference of the IEEE: 30 August - 3 September 2011; Boston, Massachusetts, USA. 2011.

13. Migeotte P, Tank J, Pattyn N, Funtova I, Baevsky R, Neyt X, et al.: Three dimensional ballistocardiography:methodology and results from microgravity and dry immersion. In Proceedings of the Engineering in Medicineand Biology Society, EMBC, 2011 Annual International Conference of the IEEE: 2011.

14. Castiglioni P, Faini A, Parati G and Di Rienzo M: Wearable seismocardiography. In Proceedings of the Engineeringin Medicine and Biology Society, 2007. EMBS 2007. 29th Annual International Conference of the IEEE: 22–26August 2007; Lyon, France. 2007.

15. Pandia K, Ravindran S, Kovacs G T, Giovangrandi L and Cole R: Chest-accelerometry for hemodynamic trendingduring valsalva-recovery. In Proceedings of the Applied Sciences in Biomedical and Communication Technologies(ISABEL), 2010 3rd International Symposium On: 2010.

16. Di Rienzo M, Meriggi P, Vaini E, Castiglioni P and Rizzo F: 24 h seismocardiogram monitoring in ambulantsubjects. In Proceedings of the Engineering in Medicine and Biology Society (EMBC), 2012 Annual InternationalConference of the IEEE: 2012.

17. Paukkunen M, Linnavuo M, Haukilehto H, Sepponen R. A system for detection of three-dimensional precordialvibrations. IJMTIE. 2012;2(1):52–66.

18. Wiens A, Etemadi M, Roy S, Klein L, Inan O: Towards Continuous, Non-Invasive Assessment of Ventricular Functionand Hemodynamics: Wearable Ballistocardiography. IEEE Journal of Biomedical and Health Informatics2014, PP(99):1.

19. Tavakolian K, Vaseghi A, Kaminska B. Improvement of ballistocardiogram processing by inclusion of respirationinformation. Physiol Meas. 2008;29(7):771–8.

20. Pandia K, Inan OT, Kovacs GTA, Giovangrandi L. Extracting respiratory information from seismocardiogram signalsacquired on the chest using a miniature accelerometer. Physiol Meas. 2012;33(10):1643–60.

21. Pan J, Tompkins WJ: A real-time QRS detection algorithm. Biomedical Engineering, IEEE Transactions on1985, (3):230–236.

22. Crow RS, Hannan P, Jacobs D, Hedquist L, Salerno DM. Relationship between seismocardiogram andechocardiogram for events in the cardiac cycle. Am J Noninvasive Cardiol. 1994;8(1):39–46.

23. Tavakolian K, Ngai B, Blaber AP and Kaminska B: Infrasonic cardiac signals: Complementary windows tocardiovascular dynamics. In Proceedings of the Engineering in Medicine and Biology Society, EMBC, 2011 AnnualInternational Conference of the IEEE: Boston, Massachusetts, USA. 2011.